Information gain expresses how much new information a document adds on top of what the system already knows from other sources. It is a defence against content that is stylistically flawless, topically correct, and at the same time completely useless, because it merely rephrases the same thing as twenty other pages. In generative search, this principle has a direct impact: the model needs to add sources to the answer that contribute something, not duplicate them. This contribution can be created in very concrete ways – with your own data and measurements, prices and limits found nowhere else, practical procedures from real cases, descriptions of exceptions and edge cases, or by updating what has changed in the meantime. The first question before writing, therefore, is not what to write about, but what is missing from existing sources.
See also: Query fan-out, Share of voice in AI responses, Grounding.
